AI Red Teaming
Prompt & Adversarial Security
- Adversarial Examples — How Noise Breaks OCR and Vision RAG: How tiny perturbations in images or text can bypass vision-based RAG pipelines.
- Bypassing LLM Guardrails: The Art of Obfuscation: Techniques to evade model safety filters and restrictions in LLMs.
- The Ultimate Prompt Injection Cheat Sheet for Red Teaming: Comprehensive guide to prompt injection attacks and defenses for testing AI systems.
- Steganography 101 for AI — The Invisible Prompt Injection: Hiding malicious instructions in text, images, or files to trick models.
File & Payload Security
- How to Craft a Polyglot File — When an Image is Also a Script: Creating files that serve as both media and executable payloads, and defending against them.